You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
EKS node route MTU is now set correctly, so container image pulls no longer silently fail on hosts with jumbo frames
EKS nodegroup teardown no longer gets stuck in DELETING after the worker nodes terminate
Nodegroup scaling and replacement no longer stall on stale readiness state from a prior nodegroup incarnation
EKS and ECS system-instance VMs now fail fast with a clear error if the guest never reaches running state, instead of hanging indefinitely
NVIDIA Blackwell GPU passthrough is now stable in EKS and ECS GPU AMI builds (VFIO binding and image generation fixed)
Compute
CreateSnapshot completes reliably — the snapshot request is now routed to the node hosting the source volume
DescribeInstances returns Placement and SpotInstanceRequestId for stopped and terminated instances, matching AWS behaviour
EC2 key pair fingerprints now match AWS exactly for both RSA and ED25519 keys
AMI import progress is now displayed in human-readable units (MB/s, GB transferred) in the console
Reliability
Instance termination no longer wedges when volumes are attached — detach is bounded with backstops for vanished QEMU processes and stuck-terminate states
Block storage chunk garbage collection is now enabled by default, reclaiming disk space from deleted and overwritten data automatically
Data volume detach completes correctly after a VM is relaunched, closing a gap that left volumes stuck attached
Volume writes no longer fail with out-of-space when object storage returns a transient SlowDown response
Block-level read-modify-write is now atomic, preventing data corruption under concurrent writers
Object storage (S3) now starts successfully on large datasets by batching FSM state restoration on startup
AWS error codes are now correctly propagated through nested error wraps, so callers receive the right error response rather than a generic one
Networking
Spinifex DNS zones are now correctly registered on hosts using systemd-resolved or resolvconf
Management IP allocation is now cluster-wide and atomic via CAS, preventing duplicate IPs from being assigned across nodes
DHCP now respects the caller's retransmission budget and encodes option 61 correctly, improving lease reliability in strict DHCP environments
